🚀 CUDA-Accelerated McEliece KEM (FYP2)

📖 Overview

This repository contains my Final Year Project 2 (FYP2) at Universiti Tunku Abdul Rahman (UTAR), focusing on GPU parallelization of the Classic McEliece post-quantum cryptosystem using CUDA.

The goal was to accelerate encryption, decryption, encapsulation, and decapsulation by offloading computationally intensive tasks (syndrome generation, error vector formation, FFT, Benes network, etc.) to NVIDIA GPUs.
Performance was benchmarked and compared with an optimised CPU vectorized implementation, showing significant throughput improvements.

✨ Features

• ✅ Full KEM flow: Encapsulation, Decapsulation, Encryption, Decryption
• ✅ CUDA kernels for FFT, Benes network, error vector generation
• ✅ Benchmarking scripts – CPU vs GPU, multiple num_blocks configurations
• ✅ CSV output for results & plotting
• ✅ Full FYP2 Report (PDF) included

🖥️ Tested Environment

Component	Specification
OS	Ubuntu 24.04.2 LTS
GPU	NVIDIA GeForce GTX 1650 (4 GB)
CUDA Toolkit	12.4
CPU	AMD Ryzen 5 3550H @ 2.10 GHz
RAM	12 GB
Driver Version	550.144.03

📊 Key Results (Throughput in Bytes/Second)

Here’s a highlight of the performance improvement:

Encrypt:

Num Blocks	CPU Encrypt (B/s)	GPU Encrypt (B/s)
1	10,917,018	2,314,525
2	13,532,555	4,791,434
4	12,782,441	9,871,298
8	12,806,684	20,305,654
16	12,542,706	41,614,452
32	12,219,651	81,403,555

Decrypt:

Num Blocks	CPU Decrypt (B/s)	GPU Decrypt (B/s)
1	2,515,560	553,932
2	2,542,354	1,165,266
4	2,517,718	2,303,712
8	2,488,658	4,657,335
16	2,418,709	9,264,162
32	2,508,264	18,515,560

📈 Observation:

• CPU throughput stays almost constant — no parallelism (loops run sequentially).
• GPU throughput rises almost linearly with num_blocks, showing strong scaling thanks to parallel execution.
• At 32 blocks, GPU achieves 6.7× faster encryption and 7.4× faster decryption compared to CPU.

⚙️ Installation & Usage

1️⃣ Clone the Repo

• git clone [https://github.com/DesmondJS/Cuda_McEliece_KEM.git]
• cd Cuda_McEliece_KEM

2️⃣ Build

• make clean
• make

3️⃣ Run

• ./run_test

🙌 Acknowledgements

This project was completed as part of UTAR FYP under the supervision of Dr. Lee Wai Kong.

📜 License

This project is licensed under the MIT License – see the LICENSE file for details.